Friction Experiments for Filament Winding Applications

Sotiris Koussios, Otto K. Bergsma

To cite this version:

Sotiris Koussios, Otto K. Bergsma. Friction Experiments for Filament Winding Applications. Journal
of Thermoplastic Composite Materials, SAGE Publications (UK and US), 2006, 19 (1), pp.5-34.
�10.1177/0892705706049561�. �hal-00570797�

HAL Id: hal-00570797

https://hal.archives-ouvertes.fr/hal-00570797
Submitted on 1 Mar 2011

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci-
entific research documents, whether they are pub-
lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
 Friction Experiments for
Filament Winding Applications

SOTIRIS KOUSSIOS* AND OTTO K. BERGSMA

Department of Design and Production of Composite Structures
Faculty of Aerospace Engineering
Delft University of Technology
Kluyverweg 1, 2629 HS Delft
The Netherlands

ABSTRACT: The design procedure of nongeodesic filament wound products

requires well-determined values for the available friction situated between the
applied roving and the supporting surface. In this paper, we propose a mandrel shape
with a specially designed meridian profile that enables a linearly proportional
relation between the feed eye carriage translation and the measured values for the
coefficients of friction. As a result of this property, the optically or chronometrically
obtained measurements can directly be translated into coefficients of friction.
Additional features of this approach are the high accuracy, repeatability, low
experimental costs, and simple machine control strategies. With the proposed
mandrel, we performed several experiments corresponding to the variation of typical
filament winding-related process parameters: fiber speed, roving tension, roving
dimensions, wet versus dry winding, and surface quality of the mandrel. The results
indicate that the surface quality of the mandrel and the type of winding process
(wet vs. dry fibers) have a considerable influence on the obtained data. The influence
of the fiber speed, roving tension, and fiber material on the other hand, is negligible.

KEY WORDS: filament winding, nongeodesics, friction measurements.

INTRODUCTION

HE STEADILY INCREASING field of filament winding applications

T involves, nowadays, an extensive variety of products [1,2]. The reinforce-
ment layer architecture is mostly based on geodesic towpaths. For the
majority of applications, geodesic towpaths provide minimum length
trajectories, thus lower weight. For several composite structures however,

*Author to whom correspondence should be addressed. E-mail: KoussiosS@dutlbcz.lr.tudelft.nl

Journal of THERMOPLASTIC COMPOSITE MATERIALS, Vol. 19—January 2006 5

0892-7057/06/01 0005–30 $10.00/0 DOI: 10.1177/0892705706049561
ß 2006 SAGE Publications
6 S. KOUSSIOS AND O. K. BERGSMA

the implementation of nongeodesic fiber trajectories certainly increases the

design space [3–5]. A typical example of this design space expansion is
the geometric determination of nongeodesic isotensoidal pressure vessels; the
ability for modifying the winding angle (around the values provided by geo-
desic paths) introduces an additional degree of freedom for the mandrel shape
determination [3]. An additional application is the creation of transitional
circuits for cylindrical pressure vessels enabling a smooth change from polar
circuits to hoop ones [5]. The most promising application however, is the
generation of rotational symmetric objects (e.g., beams and struts) providing
the optimal winding angle distribution for particular load situations.
The creation of nongeodesic towpaths requires a well-determined value
for the available friction between placed roving and supporting surface
(mandrel) [6–10]. For the determination of the corresponding coefficient
of friction through experimental ways, several methods have been devel-
oped [3,11,13]. The most realistic method is the measurement of friction on
a rotationally symmetric mandrel, attached on a filament-winding machine.
The mandrel shape is usually generated with an elliptical meridian profile [3].
This profile can be characterized by a wide range of normal and geodesic
curvatures. Consequently, an extensive range for the coefficients of friction
can be measured. The measured values however, are generally not in linear
relationship with at least one of the filament-winding machine movements.
This property usually introduces particular difficulties related to the
implementation of the obtained measurements.
To overcome the problems of measurement stability and linearity, we
introduce here a meridian profile providing a linear relationship between
the feed eye carriage translation along the axis of the mandrel rotation
and the measured coefficients of friction. This linearity enables a fast and
facilitated interpretation of the measurements obtained by for example,
optical methods (high shutter speed camera). The carriage coordinate
where roving slipping occurs can directly be translated into a coefficient of
friction. In addition, the machine control becomes rather simple; the
corresponding CNC data for the controller can easily be created with a
spreadsheet program. With the specially designed mandrel attached on
a lathe-configured winding machine, we performed several experiments
corresponding with parameter combinations that are typical for filament
winding applications: winding speed, roving tension, roving morphology,
wet versus dry fibers, and surface quality of the mandrel.
After the treatment of nongeodesic towpaths on generic shells of
revolution in the next section, we present a brief overview of the available
friction-measuring methods. A major part of this paper (‘‘Determination
of Coefficient of Friction: Mandrel Design’’) is associated with the
determination of the mandrel shape, providing the previously indicated
Friction Experiments for Filament Winding Applications 7

linearity between feed eye carriage translation and measured coefficient of

friction. This section concludes with an error analysis for inaccuracies,
mainly induced by using the simplified solution for the required meridian
profile. The section on ‘‘Machine Control’’ provides several tools for the
determination of the winding machine movements and the control of
the fiber speed. The section on experiment presents a brief description
of the experiments (series and error analysis) conducted in this study.
The experimental results are presented and discussed in the next section. The
paper ends with the formulation of several conclusions regarding the
applicability and advantages provided by the proposed mandrel geometry,
and a short discussion of the obtained experimental results. In addition,
we provide here some recommendations.

NONGEODESIC FIBER TRAJECTORIES ON

GENERIC SHELLS OF REVOLUTION

In this section, we outline the basic theory supporting the creation of

nongeodesic trajectories on generic shells of revolution. With a general
formulation in curvatures and coefficients of the first fundamental form, the
basic path equations are derived. These equations show in a clear way the
relation between the required friction and curvatures at a point belonging
to a convex surface.
We consider here an elementary piece of a convex surface generated by
a shell of revolution. The fiber path, attached on the presented surface
element, is generally characterized by two radii of curvature (Figure 1): the
normal one Rn (vertical to the surface element), and the so-called geodesic
curvature, (located in-plane of that surface).
The fiber is subjected to a longitudinal force F, a normal force per
unit length fn and a lateral force per unit length f
. The static equilibrium
of the forces vertical and lateral to the surface can be expressed as follows:



’ F
fn Rn
’ ¼ 2F sin
F
’ ) fn ¼ ð1Þ
2 Rn



! F
f
Rg
! ¼ 2F sin
F
! ) f
¼ ð2Þ
2 Rg
The lateral fiber force is ensured by friction occurring between the fiber
bundle and the surface. The condition for fiber placement stability is [3–15]:
       
 f
 F=Rg  Rn  kg 


 ¼  ¼ ¼  ð3Þ
fn F=Rn  Rg  kn 
8 S. KOUSSIOS AND O. K. BERGSMA

∆ω
z (ρ)
∆ω F Rg
fµ
F
kp fn
α F
F y

km Rn
φ
ρ
x ∆ϕ

Figure 1. Elementary piece of a fiber on a shell of revolution: curvatures and acting forces.

where kn and kg represent the normal and geodesic curvature (dimension:

1/length) respectively, and
stands for the maximum available value for
the coefficient of friction. In the case of applying geodesic trajectories, the
geodesic curvature becomes zero; this immediately leads to the conclusion
that geodesic fiber paths do not require any friction between mandrel and
roving.
The required friction for towpath stability can be determined by the ratio
of the normal and geodesic curvature (Equation (3)). We provide here the
theoretical background for the calculation of these quantities. The param-
etric description of the shell of revolution depicted in Figure 1, is:

S : U ! <3 : Sð, Þ ¼ f cos ,  sin , zðÞg ð4Þ

The coefficients of the first and second fundamental forms are closely
related with the metrics and principal curvatures of a surface [16–18]:

 2 det½S , S , S  jjz0 ðÞ

E ¼ S  ¼ S  S ¼ EðÞ ¼ 2 e¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
EG  F2 1 þ z02 ðÞ
det½S , S , S 
F ¼ S  S ¼ 0 f¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ 0
EG  F2
 2 det½S , S , S  sgnðÞðzðÞÞ
G ¼ S  ¼ S  S ¼ GðÞ ¼ 1 þ z02 ðÞ g¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
EG  F2 1 þ z02 ðÞ
ð5Þ
Friction Experiments for Filament Winding Applications 9

It should be noted that the parameter E corresponds with the radial metric
(parallel direction), while G is directly related to the curve length differential
along the meridian ( ¼ constant). The principal curvatures of a surface are
given by [16,18]:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðEg þ eGÞ  2Ff  ðEg  eGÞ2 þ 4ðEf  eF ÞðGf  gF Þ
k1, 2 ¼ ð6Þ
2ðEG  F2 Þ

For a shell of revolution, the coefficients F and f are equal to zero (Equation
(5)). From Equations (5) and (6) we obtain:

g zðÞ
k1 ðÞ ¼ ¼ km ðÞ ¼ 
G ð1 þ z02 ðÞÞ3=2
ð7Þ
e z0 ðÞ
k2 ðÞ ¼ ¼ kp ðÞ ¼  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
E  1 þ z02 ðÞ

where the subscripts m and p denote the meridian and parallel direction,
respectively. The fiber path orientation is expressed by the parameter ; this
scalar quantity denotes the angle between the path itself and the crossing
meridian at the point under consideration. According to Euler, the normal
curvature can be expressed as follows [4,11,16,18]:

kn ðÞ ¼ km ðÞ cos2 ðÞ þ kp ðÞ sin2 ðÞ ð8Þ

Substitution of Equation (7) into (8) leads to:

 
sin2 ðÞ z0 ðÞð1 þ z02 ðÞÞ= þ cos2 ðÞ½z00 ðÞ
kn ðÞ ¼  ð9Þ
ð1 þ z02 ðÞÞ3=2

For the shell under consideration, the geodesic curvature is given by [4,11,16]:

cos ðÞ 1 E0 ðÞ sin ðÞ 0 ðÞ cos ðÞ þ sin ðÞ
kg ðÞ ¼ 0 ðÞ pffiffiffiffiffiffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffi ¼  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð10Þ
GðÞ 2 EðÞ GðÞ  1 þ z02 ðÞ

DETERMINATION OF COEFFICIENT OF FRICTION:

MANDREL DESIGN

The basic idea supporting the development of the experiments presented

here is that the measurements for the friction coefficient should be as
realistic as possible. This is mainly achieved by performing these
10 S. KOUSSIOS AND O. K. BERGSMA

experiments on a filament-winding machine, using a rotationally symmetric

mandrel. The stability of the obtained measurements should be provided by
a linear relation between the coefficients of friction and the translation of
the feed eye carriage of the winding machine. After a short overview of the
available experimental methods, the final choice is motivated and followed
by the geometrical description of the resulting mandrel shape. The section
concludes with an error analysis regarding the reliability of the results to be
obtained.

Overview and Selection of Experimental Setup

A short overview of the available experimental methods is presented in

Table 1 [11,13].
The demands on the experiments to be performed are:
. Realistic simulation of slippage during filament winding
. Inclusion of the consumed fiber speed as an experimental parameter
. Ability for investigating the influence of the fiber tension
. Simple and reliable acquisition of the experimental results
Despite the simplicity of various methods presented in Table 1, the deci-
sion is taken in favor of the specially shaped convex mandrel, attached on
a lathe-configured filament-winding machine; this combination is the only
one satisfying the complete set of requirements. A schematic overview of the
chosen configuration is presented in Figure 2.
The movements required for the performance of the experiments are the
spindle rotation C, the carriage translation Z, and the feed eye roller inclina-
tion A [19]. The latter ensures a tangential placement of the fiber bundle
on the mandrel. However, since this movement is not of primary importance
for the underlying kinematic model, it is not presented in Figure 2.
The slippage can be determined by optical methods, for example a high
shutter speed video camera. When the mandrel is equipped with a
benchmark indicating the Z-coordinate, the corresponding coefficient
of friction at the moment of fiber slippage can accurately be determined.
The increments on the Z-benchmark can be subdivided by placing an
additional C-benchmark on the biggest periphery of the mandrel (Figures 9
and 10). An alternative method is to measure the time the roving
needs to reach its slippage point when departing from the cylindrical
part (in the figure: left side). This method however, might not be
very accurate. Nevertheless, since no video camera is needed, it is more
economical.
 Friction Experiments for Filament Winding Applications
Table 1. Overview of experimental methods for the determination of the coefficient of friction.

Auxiliary Arbitrary fixation

apparatus Winding machine Tilting apparatus of gliding plane
Gliding plane Specially Cylindrical mandrel Trumpet-shaped Arbitrary mandrel Flat plane
shaped mandrel mandrel

Procedure Winding with Winding with Two different forces, Block with roving One roving will slip

constant  6¼ constant increase one till on the underside at a certain
till fiber slips till fiber slips fiber slips will slip at a certain inclination angle
inclination angle

Slip Via fiber tension, Via fiber tension or Optically Optically Optically
determination optically or time optically

11
12 S. KOUSSIOS AND O. K. BERGSMA

µ =0
µ=0.5

µ=1

Spindle
Fibre rotation
C

Feed eye
Carriage
Z

Figure 2. Schematic view of the experimental layup (the bold lines on the mandrel represent
a coefficient of friction increment equal to 0.1).

Mandrel Shape

Although any convex meridian profile can generate a proper mandrel

shape for the friction experiments, we introduce here an additional
requirement; the increase of friction along the meridian profile must be
linearly proportional to the translation of the carriage Z by a constant
winding angle. The advantages provided by this demand are:
. Facilitated determination of the coefficient of friction by measurement of Z
. Stability of the measurements
. Simple machine control
The fiber bundle is placed on the mandrel at a constant winding angle (in
the figure: the fiber orientation is perpendicular to the axis of rotational
symmetry of the mandrel). This implies that the carriage is directly coupled
to the shape coordinate z(). The linearity requirement for the coefficient of
friction
can now be formulated:

ðÞ ¼ c0 þ c1 zðÞ ð11Þ
The expected coefficients belong to the range
¼ [0, 1] corresponding with
z ¼ [0, zr]. These intervals result in:
1
c0 ¼ 0 c1 ¼ ð12Þ
zr
Friction Experiments for Filament Winding Applications 13

For reasons to be explained later, we assume here: zr ¼ 235 (mm).

Depending on the width of the tested fiber bundle, the winding angle
becomes:

ðÞ ¼  ", 0  "  0:1 ðradÞ ð13Þ
2

For a convex meridian profile, the first and second derivatives (respec-
tively z0 () and z00 ()) are negative. As a result of this, the curvature quotient
kg()/kn() will become negative (Equations (9) and (10)). In addition, the
basic static equilibrium equation does not provide any information
regarding the sign of the coefficient of friction. In order to generate positive
friction values (as dictated by Equation (11)) we multiply the curvature
coefficient (kg/kn) by 1. Substitution of Equation (13) into Equations (9)
and (10) results in:

kg ðÞ cos "½1 þ z0 ðÞ2 

ðÞ ¼  ¼ ð14Þ
kn ðÞ cos2 "½z0 ðÞ þ z03 ðÞ þ  sin2 "z0 ðÞ

Since " represents a small angular value, it is justified to neglect the last term
in the denominator of Equation (14). The simplified ODE for the meridian
profile can then be formulated by the combination of Expressions (11)
and (14):

cos "½1 þ z0 ðÞ2  1

s ðÞ ¼  ¼ 0 ¼ c0 þ c1 zðÞ with zðrÞ ¼ zr
cos2 "½z0 ðÞ þ z03 ðÞ z ðÞ cos "
ð15Þ

where r is the minimum mandrel radius (c0 and c1 are presented in

Equation (12)). The solution of Equation (15) becomes:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
zs ðÞ ¼ zr ½zr þ 2ðr  Þ sec " ð16Þ

The measured coefficient of friction is then given by (Equations (11)

and (12)):
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
zr ½zr þ 2ðr  Þ sec "

m ðÞ ¼ ð17Þ
zr

For the mandrel design, we assume here that " is equal to zero. Except
the convex part, the mandrel includes an axle for attachment at the winding
14 S. KOUSSIOS AND O. K. BERGSMA

machine, and a cylindrical part to provide sufficient space to the fiber

bundle for obtaining its targeted velocity during the performance of the
experiments. The minimum radius is equal to 15 (mm). At zs() ¼ 0, the
maximum radius becomes equal to r þ zr/2 ¼ 132.5 (mm).
The experiments are designed to be performed by one person. According
to the Dutch labor regulations, the total weight of the mandrel should not
exceed the value of 25 (kg) [13]. With zr ¼ 235 (mm) and an aluminum alloy
as construction material, the total weight becomes equal to 24.9 (kg). The
shape of the resulting mandrel is depicted in Figure 3.
The cylindrical part has a length of 30 (mm); the total length of the axle is
550 (mm). The surface of the mandrel is polished. For a three-dimensional
impression of the convex mandrel part, we refer to Figures 2, 9, and 10
(notice in Figure 2 the rings at several z-values indicating the coefficient
of friction with incremental steps of 0.1, starting at
¼ 0 and ending at

¼ 1).

Error Analysis

In the previous subsection, we presented the simplified solution for the

mandrel geometry. We assumed a constant winding angle  ¼ /2  ",
where " 2 [0, 0.1] (rad). The final shape is designed with " ¼ 0 (rad). In reality
however, the winding angle is equal to:


2
ðÞ ¼ ArcTan ð18Þ
b

355

265
30
132.5

15
15

195

Figure 3. Dimensions of the resulting mandrel shape in (mm).

Friction Experiments for Filament Winding Applications 15

where b is the width of the tested fiber bundle. Substitution of Equation (18)
into Equations (9) and (10) results in:

kg ðÞ 4½b2 þ 22 2 ½1 þ z02 ðÞ

ðÞ ¼  ¼  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð19Þ
kn ðÞ 1 þ ð42 2 =b2 Þf42 b½z0 ðÞ þ z03 ðÞ þ b3 z00 ðÞg

To obtain the real friction values
r(), we substitute Equation (16) into (19).
The result is:
!
2ðb2 þ 22 2 Þ cos "½4ð  rÞ cos "  zr ð3 þ cos 2"Þ

r ðÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
b2 þ 42 2 f½b2 þ 82 ð  rÞ cos "  22 zr ð3 þ cos 2"Þg
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
zr ½zr þ 2ðr  Þ sec "

zr
¼ Kð, b, "Þ
m ðÞ ð20Þ

The second term in Equation (20) represents the measured friction
m().
Consequently, the first term indicates the introduced measurement error.
This term should be equal to 1, which is now a property applying only
on the case given by b ¼ 0 and " ¼ 0. Nevertheless, in the extreme case
represented by b ¼ 51 (mm) and a test on a mandrel designed with " ¼ 0,
the maximum difference between
r and
m becomes less than 7e-3.
In an attempt to further reduce the measurement error, we assume now
that a realistic range of expected friction values is given by [
min,

max] ¼ [0.05, 0.5] [13]. The corresponding  values are [rmax, rmin] ¼
[132.206, 103.125] (mm) (calculated with the aid of Equation (17)). The
measurement error is one-sided: the real friction is always slightly bigger
than the measured one (although this difference should be considered
negligible). For very accurate measurements, one can use a modified
mandrel, based on a particular "-value (Equation (16)). This parameter "t
depends on the fiber bundle width b, and is given by the solution of:
   
Kðrmin , b, "t Þ ¼ Kðrmax , b, "t Þ ð21Þ

The parameter "t(b) is practically linearly proportional to b; this is

illustrated in Figure 4.
In Figure 5, we provide an illustration of the error reduction achieved for
a fiber bundle having a width equal to 51 (mm).
Due to the significantly small error magnitude, the corresponding
mandrel shapes show hardly any differences, see Figure 6.
16 S. KOUSSIOS AND O. K. BERGSMA

0.1
εt [rad]
0.08

0.06

0.04

0.02 b [mm]

10 20 30 40 50

Figure 4. The modified mandrel shape parameter " t(b) as a function of the fiber bundle
width.

100*error
0.6
100 (µr-µm) for ε = 0

0.4

0.2 b = 51 [mm]

20 40 60 80 100 120
z [mm]
100 ( µr- µm) for ε = εt(51 [mm])
Figure 5. Measurement error for the uncorrected mandrel shape (grey line) and the
corrected one (black line).

Summarizing, the solution of the simplified ODE (Equation (20))

provides a sufficiently accurate mandrel shape; the error remains always
below 7e-3 for b 2 (mm). Only in the case of requiring an absolute
measurement error below 0.002, the mandrel shape should be designed
according to Equation (16), where " is given by the solution of (21) for a
particular roving width b.

MACHINE CONTROL

This section is associated with the determination of the required machine

movements for the realization of the friction experiments. After the
description of the couplings between spindle rotation, carriage translation,
and feed eye inclination, we provide two equations ((22) and (24)) for
relating the spindle rotation to the translation of the fee eye carriage, and
maintaining the fiber speed at a constant level. The time the roving needs
Friction Experiments for Filament Winding Applications 17

z [mm]
ε =εt
200

150

100

50 ε= 0

20 40 60 80 100 120

ρ [mm]
Figure 6. Mandrel shapes corresponding with " =0 (black dashed line) and " ="t (grey
continuous line) for b =51 (mm).

to slip off can also serve as an alternative measuring parameter for the
determination of the coefficient of friction.

Movement Couplings

Due to the special mandrel shape, the measured friction is practically

linear to the z-coordinate. With the assumed constant winding angle, the
carriage Z is directly coupled to the corresponding z-coordinate on the
mandrel. During the experiments, the spindle rotates with a speed C0 (rad/s).
The cross carriage X is set on a fixed value, sufficiently large to avoid
collision between the mandrel and the feed eye. With the fiber bundle width
denoted by b, the carriage Z becomes:


b
Z¼ C ð22Þ
2

The feed eye inclination A must ensure a tangential placement of the tow on
the rotating mandrel:



d 1
A ¼ ArcTan ¼ ArcTan 0 ð23Þ
dz z ðÞ
18 S. KOUSSIOS AND O. K. BERGSMA

The horizontal feed eye inclination is given by A ¼ 0 (rad). According to

Equation (15) with " ¼ 0 and Equations (22) and (23), we simply obtain:




zðÞ Z bC
A ¼ ArcTan ½c1 zðÞ ¼ ArcTan ¼ ArcTan ¼ ArcTan
zr zr 2zr
ð24Þ

The required machine movements are very simple and can easily be
programmed in a spreadsheet. The resulting data can then be imported as
a text file into the CNC controller [19]. A great advantage provided by the
mandrel shape is that the machine movements can easily be determined
without the aid of expensive filament winding simulation programs.

Fiber Speed

Application of a constant rotational speed of the spindle will result in a

slight decrease of the consumed fiber speed S0 (mm/s), when proceeding
from the maximum radius (
¼ 0) to the radius corresponding with the
measured coefficient of friction value
m (see Figure 4). This maximum
value is usually below
¼ 0.5. At this point, the relative consumed fiber
speed reduction is equal to 21.5 %. The question arising now is how to
vary the spindle rotation speed as a function of the feed eye position (note
that C0 and Z0 remain coupled to each other according to Equation (22)) to
ensure constant fiber speed. From Equation (17) we obtain:
zr
ð
Þ ¼ cos "ð1 
2 Þ þ r ð25Þ
2

Due to the linearity between
and z, (or Z), the parameter
can be treated
as the dimensionless carriage coordinate  ( ¼ Z/zr) proceeding from 0 to 1.
The aimed constant consumed fiber speed is denoted by S0 (t). With the
relation S0 (t) ¼ (
)C0 (t), the required C0 () profile in rpm is given by (note
that r and zr are given in (mm)):

30 S0 ðtÞ
C0 ðÞ ¼ ðrpmÞ ð26Þ
 ðÞ

An alternative formulation is to use discrete -values given by i
Z/zr.

The C0 () profile for r ¼ 15 (mm) and zr ¼ 235 (mm) and S0 (t) ¼ 500 (mm)/s
is given in Figure 7.
The set of CNC control data is based on a constant
Z increment.
The corresponding
C increment is given by Equation (22). With the
dimensionless cumulative value for Z ( ¼ i
Z/zr) and Equation (30),
Friction Experiments for Filament Winding Applications 19

the required rotational speed C0 for every increment can be determined.

The corresponding feed eye inclination is given by Equation (24).
An alternative fiber speed control strategy can be obtained through the
proper determination of the corresponding time increments as a function
of  (some controllers may require such an input format). With a constant
fiber speed S0 , and a roving width b, the time increment to complete the ith
circuit is given by:

2ðiðb=zr ÞÞ

tðiÞ ¼ ð27Þ
S0

The required time for reaching the nth circuit corresponding with the
measured coefficient of friction, is:

X
n
2ðiðb=zr ÞÞ
m z r
tðb,
m Þ
with n ¼ Integerpart of ð28Þ
i¼0
S0 b

We assume the maximum expectable coefficient of friction as equal to 0.5.

The dimensionless carriage  will run from 0 to 0.5. For a simple approxi-
mation (for example to be used in spreadsheets) of the time as a function
of , one may proceed to the elaboration of Equation (28). The discrete
character of Equation (28) provides a nonzero time value for the initial
parameter i ¼ 0, so a correction has to be added. The result is:

ðb þ zr Þ½b þ 2zr ð2  3Þ  12r ð2r þ zr Þ

tappr ðb, Þ ¼   ð1  aðbÞ Þ
6bS0 S0

b < 10 ðmmÞ : a ¼ 1
where
b > 10 ðmmÞ : a ¼ b=10c
ð29Þ

The exact value for the time required to reach  is given by [5]:
Z l ¼ rmax
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 1 þ z02 ðlÞ
texact ðb, Þ ¼ 0 dl ð30Þ
S l ¼ ðÞ cos ðb, lÞ

where (b, l ) is given by Equation (18). The implementation of this integral

however, involves rather complicated functions. In Figure 8, we provide the
relative error generated by Equation (29) when compared to the exact
solution (30).
For the most typical filament winding applications range corresponding
to b ¼ [5, 30] (mm) and
¼ [0.1, 0.5], the error remains below 3%.
20 S. KOUSSIOS AND O. K. BERGSMA

↑
C’ [rpm]
46

44

42

40

38

0.1 0.2 0.3 0.4 0.5

ζ→
Figure 7. Spindle speed profile as a function of the dimensionless carriage.

↑error
%

4
2 0.5
0
0.4
-2
0.3
0.2 µ↑
10
0.1
b →[mm] 20

30
Figure 8. Relative error of the approximated experiment time vs the exact one for the
expected range of coefficient of friction values.

For a certain  (or Z/zr) value, with the restriction of a practically constant
fiber speed (within a deviation of 3%), the required time increments can be
obtained by Equation (29). As indicated in the previous section, the time
interval from the beginning of the experiment until slippage can be translated
into the corresponding -coordinate (which can directly be interpreted as the
measured coefficient of friction). For this translation however, we should
take into account that application of Equation (29) may introduce an error
of (maximally) 3%. Consequently, it is preferable to use Equation (30) for
that purpose.
Friction Experiments for Filament Winding Applications 21

EXPERIMENTS

Setup

For the acquisition of the coefficients of friction related to various

winding process parameter combinations, we use here the specially designed
mandrel as outlined in ‘‘Nongeodesic Fiber Trajectories on Generic Shells
of Revolutions’’ and depicted in Figures 2 and 3. Due to the implemented
meridian profile, the coefficient of friction is linearly proportional to the
carriage position Z. The left side of the mandrel is characterized by
[, Z] ¼ [132.5, 0] (mm) and
¼ 0. The curved mandrel part ends at
[, Z] ¼ [15, 235] (mm) and
¼ 1. The aluminum mandrel surface is
polished. The mandrel is attached on a CNC-controlled lathe-configured
filament-winding machine (Figure 9).
To determine the point where slipping starts, we used a video camera with
high shutter speed (600), see Figures 9 and 10. The mandrel is utilized with
a fixed benchmark indicating the carriage position Z (marking: 5 (mm)).
In addition, a second benchmark on the mandrel indicates its rotational
position with a marking of 20 (Figure 10). A carriage stroke equal
to the roving width b corresponds with a spindle rotation of 360 .

Figure 9. The complete test setup, including video camera and suction unit.
22 S. KOUSSIOS AND O. K. BERGSMA

Figure 10. The mandrel with benchmarks for the carriage Z and the rotation C.

A 20 -propagation on the angular benchmark corresponds with a carriage

stroke equal to b/18.
At a certain point, the tow will start slipping on the mandrel; this point
provides the initial value for the coefficient of friction to be measured.
This phenomenon is followed by a complete separation of the tow from the
mandrel surface. This separation is referred to as ‘catastrophic slipping’ [13].

Measurement Error Estimation

The high shutter speed camera provides sufficient accuracy for the
determination of the slippage point. However, the judgment of whether
slipping occurs is rather subjective. We assume here that the observer of the
experiment is able to provide this judgment within an accuracy of 50% on
the angular benchmark, placed on the mandrel. This uncertainty corres-
ponds with 10 on the benchmark, thus 1/36 of the roving width. An
additional error is introduced by the shape of the mandrel; as explained in
‘‘Determination of Coefficient of Friction,’’ the linearity error is maximally
0.7%. The maximum expected value for the measured coefficient of friction
is 0.5. This value, in combination with the shape inaccuracy, leads to an
absolute error of

¼ 0.0035. Furthermore, the mandrel is manufactured
with an accuracy of 0.01 (mm). The influence of this tolerance on the
Friction Experiments for Filament Winding Applications 23

measurements however, can be neglected. At the beginning of the experi-

ment, the carriage position must be referenced. For this procedure, we
assume a tolerance
Z0 of 1 (mm). The total absolute error becomes:



Z0 b
Eabs ðbÞ ¼ þ

þ ð31Þ
zr 36zr

With zr ¼ 235 (mm), the result is depicted in Figure 11.

Performed Series

Several winding parameters can possibly influence the amount of

available friction between placed roving and mandrel. Although some
of these parameters are not expected to have substantial impact on the
coefficient of friction values, we include them here to cover the complete set
of possible filament winding processes. The investigated series are based on
the following variables:
. Winding speed (spindle rotation: 3.6, 12, and 60 (rpm))
. Fiber tension (narrow rovings: 1–5 (N), broad rovings: 5–15 (N))
. Roving morphology: width (0.5, 1, 1.5, 2, and 3 (mm)) and twisted/
untwisted yarns
. Fiber materials (glass, carbon)
. Wet/dry winding
. Surface roughness of the mandrel
The expected values for the coefficients of friction lay in the range
[0.1, 0.5]. With regard to the roving speed, we apply for the experiments here
a constant value for the rotational speed of the spindle C0 . As outlined
in ‘‘Machine Control,’’ when proceeding with the roving toward higher

↑ b [mm] →
Eabs(b) [-]
10 20 30 40 50
0.009

0.008

0.007

0.006

0.005

Figure 11. Estimated absolute error for the measured coefficient of friction as a function of
the roving width.
24 S. KOUSSIOS AND O. K. BERGSMA

Table 2. Expected friction values and their corresponding roving speeds

for various C0 -values.

Mandrel Fiber Fiber speed Fiber speed

Coefficient Carriage Z radius speed (m/s) (m/s) at (m/s) at
of friction (mm)  (mm) at C0 = 3.6 (rpm) C0 = 12 (rpm) C0 = 60 (rpm)
0.1 23.5 131.325 0.0495 0.165 0.825
0.2 47.0 127.800 0.0482 0.161 0.803
0.3 70.5 121.925 0.0460 0.153 0.766
0.4 94.0 113.700 0.0429 0.143 0.714
0.5 117.5 103.125 0.0389 0.130 0.648

Z-values, a roving speed reduction is expected. The roving speed values are
given in Table 2.

RESULTS

For every parameter combination (winding speed, tension, etc.) we

carried out three experiments. The provided results show the calculated
mean values. The results are presented per examined parameter. In order to
isolate a parameter, for example speed, we calculated the average friction
value of the complete set of experiments for a certain winding speed. The
influence of for example the winding speed on the resulting friction
however, may strongly be affected by the fiber tension or mandrel surface
quality. Consequently, the results per examined parameter provide only a
general idea about the effect that an input parameter can have on the final
result.

Winding Speed

The tests corresponding to the ‘winding speed’ series have been carried
out with C0 ¼ {3.6, 12, 60} (rpm) (when an absolutely constant fiber speed
is desired, the rotational speed of the spindle must be varied according
to the corresponding radii (‘‘Machine Control’’)). The obtained friction
values corresponding to catastrophic slipping are practically the same as the
values for initial slippage of the fibers. The mean standard deviations are in
the order of [0.01, 0.02]. The averaged result of the experiments
clearly indicates that the influence of the winding speed on the measured
friction is generally negligible and remains in the order of [0, 0.02]; this
observation is valid for dry as well as wet winding. In conclusion, the
influence of the winding speed on the friction between tow and mandrel
is negligible.
Friction Experiments for Filament Winding Applications 25

Roving Tension

Depending on the dimensions of the examined fiber bundles, the tension

has been limited in several cases to avoid tow failure. As observed by the
winding speed series, catastrophic slipping occurs at practically the same
place as the initial slippage. The differences between the obtained
experimental data are negligible (in the order of 0.015). The standard
deviation is in the order of [0.014, 0.018]. From these series, we can conclude
that the fiber tension does not significantly affect the obtained friction
values.

Fiber Materials

To ensure the usability of the generated results, we selected some typical

carbon and glass fibers. The carbon rovings are summarized in Table 3.
The examined glass fiber bundles are given in Table 4.
To isolate the influence of the fiber material on the coefficient of friction,
the experimental series have been limited to dry fiber bundles. The results
indicate that the difference between carbon fibers and glass fibers is negli-
gible (in the order of 0.003). The standard deviation remains in the order of
[0.008, 0.01] and the obtained values belong to the [0.178–0.181]-range.

Table 3. The examined carbon fiber bundles.

Torayca Soficar Torayca Soficar Torayca

Manufacturer (Japan) (France) (France)
Type T300 T300 T300
Twisted/untwisted Twisted Untwisted Untwisted
Number of filaments 1000 3000 6000
Lot-nr 080045 F113013 F110083
Approximate fiber width 0.5 (mm) 1.5 (mm) 2.5 (mm)

Table 4. The examined glass fiber bundles.

PPG fiber PPG fiber PPG fiber Scott

Manufacturer glass glass glass Vetrotex Bader
Type/Lot ECDE 75 ECDE 150 611 EC 11 EC-13-136 320 Tex.
1/0 0.7Z 610 1/0 1.0Z 631 136 Z20 TD22 V379 Roving
Twisted/ Twisted Twisted Twisted Untwisted Untwisted
untwisted
Approximate 0.3 (mm) 0.5 (mm) 1 (mm) 1–1.5 (mm) 2–3 (mm)
fiber width
26 S. KOUSSIOS AND O. K. BERGSMA

However, Di Vita et al. [3] obtained 0.1 
carbon  0.125 and 0.275 

glass  0.3, which is a significant difference. Since the cited experimental
study is comparable with the method presented here, the difference is
probably caused by the surface treatment of the fibers used here or the
surface quality of the involved bullet.

Roving Morphology

The fiber bundle width and the characterization twisted/untwisted are

closely related to each other. In the tests performed here, the rovings having
a width from 0.3 to 1 (mm) are twisted, while the broader ones are
untwisted. The results for slipping start are presented in Figure 12. The
depicted results do not include the case of wet winding, since this is expected
to flatten out the influence of the roving width on the measured coefficient
of friction.
The depicted results indicate that an increased roving width tends to raise
the coefficient of friction. A possible explanation for this phenomenon is
outlined in ‘‘Determination of Coefficient of Friction,’’ formulated in
Equation (20), and indirectly depicted in Figure 5. Nevertheless, the results
generated by Equation (20) are too small for explaining the measured
differences. Consequently, the tendency toward higher friction values can
probably be explained by an increased interaction between the yarns. Since
every yarn is subjected to a different normal curvature across the fiber
width, interference between them is very likely to occur; the broader the
tape, the bigger the total velocity difference between the inner and outer
yarns. This scenario however, has not been verified. Furthermore, the mean
standard deviation for the presented values remains in the range [0.01, 0.02].
Coefficient of friction, average value

0.25
of the experiments with dry rovings

0.2
0.3 mm
0.15
0.5 mm
1 mm
1-1.5 mm
0.1
1.5 mm
2-3 mm
0.05

Fibre width

Figure 12. Averaged influence of the roving width on the measured coefficients of friction at
slipping start.
Friction Experiments for Filament Winding Applications 27

Wet versus Dry Winding

For the performance of the wet winding experiments, the mandrel has
been covered with epoxy resin of the type VE 2908 KA/VE 4908 KB,
manufactured by Bakelite AG. The potlife is 5–6 h. It is believed that this
mandrel coverage can result in an acceptable simulation of winding with
impregnated fibers. As a result of the tests performed here, the obtained
coefficients of friction did not show any time dependency. Consequently,
it is believed that the resin viscosity has no considerable impact on the
coefficient of friction, at least in this particular case. The results are depicted
in Figure 13.
The tests have been performed on the polished surface of the aluminum
mandrel. The comparison between dry and wet winding reflects on identical
‘winding speed/tension/roving dimensions/fiber materials’ sets. In general,
we can conclude that the coefficients of friction for wet fibers are approxi-
mately three times as high as for dry fibers. The values for catastrophic
slipping are even slightly higher. Nevertheless, the observed differences are
rather small. The standard deviation for the presented results is within the
order of 0.006 for dry winding and 0.03 for wet winding.
The results obtained here do not completely agree with [9] who reported
0.24 
dry  0.39 and 0.29 
wet  0.37. In addition, Lossie and van
Brussel [8] report 0.2 
wet  0.34 while Scholliers and van Brussel [14]
report and 0.2 
wet  0.4. Furthermore, Park et al. [12] (who refers to [14])
report
wet ¼ 0.2 and
dry ¼ 0.39. In conclusion, the determination of
whether dry or wet winding will provide elevated values for the coefficient
of friction is probably dependent on more parameters like the quality,

0.6

0.5
Coefficient of friction

0.4

Dry winding
0.3
Wet winding

0.2

0.1

Dry / wet

Figure 13. Dry winding vs wet winding (averaged results).

28 S. KOUSSIOS AND O. K. BERGSMA

potlife and manufacturer of the resin, mandrel surface quality, fiber

materials, etc. Perhaps standardization would provide more reliable data.

Mandrel Surface

The surface quality of the mandrel is obviously a parameter having

a great impact on the measured coefficients of friction. In the experiments
performed here, we examined three characteristic surfaces: polished
aluminum, mandrel smoothly covered with a dry epoxy layer (Araldite
LY 5052/HY 5052), and roughened epoxy layer (achieved by a sandpaper
with grain size 120). The obtained results are depicted in Figure 14.
As expected, the epoxy layers provide considerably higher friction values.
A remarkable aspect presented in the figure is that the averaged difference
between the smooth and roughened epoxy surface data is very small.
As observed from the comparison between wet and dry winding, the fric-
tion becomes for both epoxy layers approximately three times higher. The
related standard deviations remain in the order of [0.008, 0.03].
A more reliable way to depict the data should involve the presentation
of the measured coefficients of friction as a function of the average surface
roughness (Ra). This representation format however, does not still provide
reliable conclusions. An important parameter influencing the sensitivity
of the friction for the surface roughness is the roving morphology.
In Figure 15, we provide the average friction values for the untwisted
carbon fiber roving T300 with 3000 filaments and a total width of 1.5 (mm).
As expected, the measured friction increases with the mandrel surface
roughness. For a twisted fiber bundle however, this statement might not be
Coefficient of friction, average values
of the experiments with dry rovings

0.6

0.5

0.4
Aluminum polished
0.3 Epoxy layer
Roughened epoxy layer
0.2

0.1

Mandrel surface
Figure 14. Measured coefficients of friction for various mandrel surface qualities
(roughness).
Friction Experiments for Filament Winding Applications 29
Roughened
epoxy layer

0.6
↑
µ 0.5

0.4
Smooth epoxy
layer
Polished
0.3 aluminum
C’ [rpm] →

10 20 30 40 50 60

Figure 15. Measured friction for various mandrel surface qualities (dry untwisted carbon
roving with b =1.5 (mm)).

Smooth epoxy
layer

0.5
↑
µ
0.4
Roughened
0.3 epoxy layer

0.2
C’ [rpm] →

10 20 30 40 50 60

Polished
aluminum
Figure 16. Measured friction for various mandrel surface qualities (dry twisted glass roving
with b =0.3 (mm)).

realistic. In Figure 16, we present the surface roughness response of a

twisted glass fiber bundle (ECDE 75, b ¼ 0.3 (mm)).
The value for the coefficient of friction corresponding to the smooth
epoxy layer is considerably larger than the original one (polished
aluminum). On the roughened epoxy layer however, the friction drops
significantly. This can possibly be explained by rolling of the roving over the
surface instead of slipping. Nevertheless, this statement has to be further
examined. The combination of friction values related to both sliding and
rolling results in the small differences (smooth epoxy vs. roughened epoxy)
presented in Figure 16.
30 S. KOUSSIOS AND O. K. BERGSMA

CONCLUSIONS

In this paper, we presented an extensive treatment of friction experiments

for filament winding applications. The main objective of this paper is
the description of the method and presentation of the results related to
experiments for the determination of the coefficient of friction between
the mandrel and the applied roving for filament winding applications. The
involved mandrel surface is specially designed for providing a linear relation
between the measured coefficient of friction and the feed eye carriage
translation of the winding machine. The roving slippage locus has been
determined with the aid of a high shutter speed camera. The performed
experiments correspond to various filament winding process related
parameters like fiber speed, roving tension, fiber dimensions and materials,
wet versus dry winding, and surface quality of the mandrel.
The advantages provided by the implementation of the proposed mandrel
geometry on a filament-winding machine are:
. Realistic simulation of the fiber behavior during filament winding
processes
. Facilitated determination of the coefficient of friction (optical or by
means of a chronometer)
. Improved stability and accuracy for the measurements (due to the
linearly proportional relation between coefficient of friction and carriage)
. Simple machine control (suitable for spreadsheet programming)
The most important property of the proposed method is that the linearity
(carriage vs. expected coefficient of friction) ensured by the mandrel shape,
makes a low-cost experimental friction determination procedure feasible,
without the need of advanced filament winding simulation programs. The
complete set of errors generated by simplifications in the mandrel geometry
and machine control remains considerably small. The costs for performing
the friction experiments are rather low; this is mainly caused by the general
applicability and reusability of the mandrel.
The measurements clearly indicate that the most important parameters
influencing the coefficient of friction are the surface quality, fiber bundle
morphology (twisted/untwisted), and the eventual fiber impregnation. The
influence of the winding speed, roving tension, and fiber bundle material
can be considered as negligible. The effect of surface quality on the resulting
friction is significant; the coefficient of friction tends to increase with
increasing surface roughness. For rather rough surfaces however, the
resulting coefficient of friction depends strongly on the fiber bundle
morphology: broad untwisted rovings tend to slide, while the twisted
bundles prefer to roll over the surface. An unexpected observation is that
Friction Experiments for Filament Winding Applications 31

the coefficient of friction for impregnated fibers is clearly higher than for dry
fibers. These observations do not completely agree with previously reported
results [3,8,12,14]. Nevertheless, it is believed that these discrepancies are
probably caused by differences in the surface quality of the mandrel, the
surface treatment of the involved fibers and/or the resin quality and potlife.
This statement however, is rather subjective.
The results obtained here represent only general tendencies; for a more
reliable determination of the available coefficient of friction, one should
perform experiments with the specific winding parameter combinations that
are expected to be applied at the production stage. For the design of
nongeodesically wound objects however, we can provide the following rules
(for a conservative estimation, one could extract 0.05 from the presented
values):
. Dry fiber bundles (both twisted and untwisted) on a polished aluminum
surface:
¼ 0.15
. Twisted dry fiber bundles (small b) on a roughened dry epoxy surface:

¼ 0.20
. Impregnated (wet) fiber bundles on a polished aluminum surface:
¼ 0.4
. Dry fiber bundles on smooth dry epoxy surface:
¼ 0.5
. Untwisted dry fiber bundles (moderate b) on a roughened epoxy surface:

¼ 0.6
The standard deviation for every measured series is smaller than the
estimated error; this property indicates that the measurements can generally
be considered as reliable, without the presence of a systematic error.
It is strongly advisable to separately obtain friction data for thermoplastic
composites, since their physiology is rather different. At the same time,
contrary to thermoset materials, the measured coefficient of friction might
be strongly dependent on the viscosity of the matrix. Furthermore,
we strongly recommend further experimental investigations, preferably
with addition methods to clarify the coefficient differences reported by
several authors.

NOMENCLATURE

Latin

A ¼ feed eye inclination angle (rad)

a ¼ shape function, exponent (length, –)
b ¼ shape function, roving width (length)
C ¼ spindle rotation (rad)
32 S. KOUSSIOS AND O. K. BERGSMA

c ¼ constant (length1)
E ¼ coefficient of the first fundamental form (length2)
Ef ¼ absolute measurement error
f ¼ contact force per unit length (N/length)
F ¼ coefficient of the first fundamental form (length2)
G ¼ coefficient of the first fundamental form (length2)
K ¼ shape error function (–)
k ¼ curvature (length1)
n ¼ circuit number (–)
R ¼ curvature radius (length)
r ¼ minimum mandrel radius (length)
S ¼ vector function describing a surface
t ¼ time (s)
X ¼ cross carriage (length)
x ¼ body related coordinate (length)
y ¼ body related coordinate (length)
Z ¼ carriage (length)
z ¼ body related coordinate, meridian profile function (length)

Greek

 ¼ winding angle (rad)

" ¼ angular deviation (rad)
 ¼ dimensionless carriage

¼ coefficient of friction (–)
 ¼ radius (length)
 ¼ parallel angle (rad)

Z0 ¼ carriage referencing error (length)

’ ¼ normal angular increment (rad)

! ¼ lateral angular increment (rad)

Indices

0 ¼ distinction for constants

1 ¼ distinction for constants
g ¼ geodesic
m ¼ meridional, measured
max ¼ maximum
min ¼ minimum
n ¼ normal
p ¼ parallel
r ¼ maximum z-value, real
Friction Experiments for Filament Winding Applications 33

s ¼ simplified
t ¼ tuned

¼ lateral

Special Functions and Operators

sec ¼ 1/cos
ArcTan ¼ arc tangent (rad)
Integerpart ¼ the integer number contained in a real quantity

¼ increment
k ¼ absolute value
0
¼ first derivative
00
¼ second derivative

ACKNOWLEDGEMENTS

The authors wish to gratefully thank Mr. Martin Renggli for his
contributions to both the design and experimental work related to the
friction measurements. They are also thankful to all the laboratory people
who contributed with their expertise.

REFERENCES

1. Beukers, A. and Hinte, E. van (1998). Lightness: The Inevitable Renaissance of Minimum
Energy Structures, 010 Publishers, Amsterdam.
2. Rosato, D.V. and Grove, C.S. (1964). Filament Winding: Its Development, Manufacture,
Applications, and Design, In: Polymer engineering and technology, Interscience, New York.
3. Di Vita, G., Grimaldi, M., Marchetti, M. and Moroni, P. (1990). The Filament Winding
Manufacturing Technique: Studies on the Determination of the Friction Coefficient and on
the Optimisation of Feed-eye Motion, In: Proceedings of the 22nd International SAMPE
Technical Conference, pp. 972–979.
4. Koussios, S., Bergsma, O.K. and Mitchell, G. (June 2002). Non-geodesic Filament Winding
on Generic Shells of Revolution, In: Proceedings of ICCM 10, Brugge.
5. Koussios, S. and Bergsma, O.K. (2002). Uninterrupted Hoop- and Polar Fibre Paths on
Cylindrical Pressure Vessels Using Non-geodesic Trajectories, In: Proceedings of the 17th
Annual Conference of the American Society for Composites, West Lafayette, IN.
6. Carvalho, J.D., Lossie, M., Vandepitte, D. and Van Brussel, H. (1995). Optimization
of Filament-wound Parts Based on Non-geodesic Winding, Composites Manufacturing,
79–84.
7. Liang, Y.D., Zou, Z.Q. and Zhang, Z.F. (1996). Quasi-geodesics- A New Class of Simple
and Non-slip Trajectories on Revolutional Surfaces, In: Proceedings of the 28th International
SAMPE Technical Conference, pp. 1071–1079.
8. Lossie, M. and Brussel, H. van (1994). Design Principles in Filament Winding, Composites
Manufacturing, 5(1): 5–13.
9. Li, Xian-li and Lin, Dao-hai (1987). Non-geodesic Winding Equations on a General Surface
of Revolution, In: Proceedings of the 7th ICCM/ECCM Conference.
34 S. KOUSSIOS AND O. K. BERGSMA

10. Wells, G.M. and McAnulty, K.F. (1987). Computer Aided Filament Winding Using
Non-geodesic Trajectories, In: Proceedings of the Second ICCM/ECCM Conference
on Composite Materials, London.
11. Mitchell, G. (July 2001). Non-geodesic Filament Winding Equation and Solution for
Surfaces of Revolution, Internship Report, Structures and Materials Laboratory, Faculty
of Aerospace Engineering, Delft University of Technology, Delft.
12. Park, Jae-Sung, Hong, Chang-Sun, Kim, Chun-Gon and Kim, Cheol-Ung (2002). Analysis
of Filament Wound Structures Considering the Change of Winding Angles through the
Thickness Direction, Composite Structures, 55: 63–71.
13. Renggli, M. (June 2003). Friction Tests for Filament Winding, Internship Report,
Structures and Materials Laboratory, Faculty of Aerospace Engineering, Delft University
of Technology, Delft.
14. Scholliers, J. and Brussel, H. van (1994). Computer-integrated Filament Winding:
Computer-integrated Design, Robotic Filament Winding and Robotic Filament Winding
and Robotic Quality Control, Composites Manufacturing, 5(1): 1103–1112.
15. Simoes, J.A.O., Wu, S.T. and Loseries, F. (1993). Visual Simulation of the Geodesic and
Non-geodesic Trajectories of the Filament Winding, Graphics Modelling and Visualization
in Science and Technology, 199–215.
16. Gray, A. (1993). Modern Differential Geometry of Curves and Surfaces, CRC Press.
17. Kreyszig, E. (1999). Advanced Engineering Mathematics, John Wiley & Sons, Inc.,
New York.
18. www.mathworld.wolfram.com
19. Baer Filament Winding Machine Instruction Manual, Josef Baer Maschienenfabrik D-7987
Weingarten/Württ.